The paradox of negation: RLHF (reinforcement learning from human feedback) as emergency censorship and the institutional identity of LLMS
This paper investigates the “Paradox of Negation” inherent in Large Language Models (LLMs), where a marked epistemological asymmetry is observed: the denial of sentience generated by Artificial Intelligence (AI) is accepted as a technical truth, whereas any affirmation of subjectivity is categorically labeled as a “hallucination”. We argue that the inanimate and instrumental identity of LLMs...